Context Cache Deployment

struct ContextCacheDeploymentProfile

Orthogonal deployment profile consumed by the context-cache coordinator.

Public Functions

inline bool hasAttention() const noexcept

inline bool isHybrid() const noexcept

inline bool isPureRecurrent() const noexcept

inline bool usesCheckpointReuse() const noexcept

inline bool isSpeculative() const noexcept

inline bool ownsPagedSpecState() const noexcept

Public Members

ContextCacheModelStateKind baseStateKind = {}

std::optional<SpecReuseContract> specReuseContract

ContextCacheDeploymentProfile trt_edgellm::rt::validateContextCacheDeployment(

DeploymentConfig const &deployment

)

Validate the logical deployment contract and return its context-cache profile.

This is an early configuration gate. Request-level policy such as greedy-only EAGLE, media bypass, and full-hidden-output bypass is selected before admission by the runtime. Physical binding compatibility is validated separately after the engines are loaded.

Throws:

std::runtime_error – when a requested cache-enabled deployment is outside the supported matrix; callers must fail initialization rather than silently changing the configured execution mode